The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages the exposure of Meteor methods to external interfaces through packages like wreiske:meteor-wormhole and meteor-rpc. These tools can expose internal functionality to Model Context Protocol (MCP) agents or REST clients, creating a surface for indirect prompt injection if untrusted inputs are not handled correctly.
Ingestion points: External data enters via the /mcp and /api/ endpoints defined in references/wormhole.md, and through typed RPC calls described in references/meteor-rpc.md.
Boundary markers: The skill requires an explicit opt-in configuration, mandatory authentication, and authorization checks on the server for all exposed methods.
Capability inventory: Methods exposed via these packages can perform database writes (Mongo), trigger transactions, and interact with external services.
Sanitization: The instructions mandate strict schema validation using Zod, check, or other validator libraries to sanitize incoming arguments before they are processed by the server.
[EXTERNAL_DOWNLOADS]: The skill recommends several third-party packages from the Atmosphere and npm registries to provide features like soft deletion, archiving, and RPC integration.
Evidence: References packages including jam:method, jam:pub-sub, dupontbertrand:cluster, and meteor-rpc. The documentation links to their respective GitHub repositories and emphasizes verifying maintenance signals and version compatibility.

meteor-community-packages